Результаты исследований: Публикации в книгах, отчётах, сборниках, трудах конференций › статья в сборнике материалов конференции › научная › Рецензирование
Comparison of Reinforcement Learning Based Control Algorithms for One Autonomous Driving Problem. / Kabanov, Stepan; Mitiai, German; Wu, Haitao; Petrosian, Ovanes.
Mathematical Optimization Theory and Operations Research: Recent Trends - 21st International Conference, MOTOR 2022, Revised Selected Papers. ред. / Yury Kochetov; Anton Eremeev; Oleg Khamisov; Anna Rettieva. Springer Nature, 2022. стр. 338-349 (Communications in Computer and Information Science; Том 1661 CCIS).Результаты исследований: Публикации в книгах, отчётах, сборниках, трудах конференций › статья в сборнике материалов конференции › научная › Рецензирование
}
TY - GEN
T1 - Comparison of Reinforcement Learning Based Control Algorithms for One Autonomous Driving Problem
AU - Kabanov, Stepan
AU - Mitiai, German
AU - Wu, Haitao
AU - Petrosian, Ovanes
N1 - Publisher Copyright: © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Autonomous driving systems include modules of several levels. Thanks to deep learning architectures at the moment technologies in most of the levels have high accuracy. It is important to notice that currently in autonomous driving systems for many tasks classical methods of supervised learning are no longer applicable. In this paper we are interested in a specific problem, that is to control a car to move along a given reference trajectory using reinforcement learning algorithms. In control theory, this problem is called an optimal control problem for moving along the reference trajectory. Airsim environment is used to simulate a moving car for a fixed period of time without obstacles. The purpose of our research is to determine the best reinforcement learning algorithm for a formulated problem among state-of-the-art algorithms such as DDPG, PPO, SAC, DQN and others. As a result of the conducted training and testing, it was revealed that the best algorithm for this problem is A2C.
AB - Autonomous driving systems include modules of several levels. Thanks to deep learning architectures at the moment technologies in most of the levels have high accuracy. It is important to notice that currently in autonomous driving systems for many tasks classical methods of supervised learning are no longer applicable. In this paper we are interested in a specific problem, that is to control a car to move along a given reference trajectory using reinforcement learning algorithms. In control theory, this problem is called an optimal control problem for moving along the reference trajectory. Airsim environment is used to simulate a moving car for a fixed period of time without obstacles. The purpose of our research is to determine the best reinforcement learning algorithm for a formulated problem among state-of-the-art algorithms such as DDPG, PPO, SAC, DQN and others. As a result of the conducted training and testing, it was revealed that the best algorithm for this problem is A2C.
KW - Autonomous driving
KW - Control algorithms
KW - Optimal control of moving along the reference trajectory
KW - Reinforcement learning
UR - http://www.scopus.com/inward/record.url?scp=85140465648&partnerID=8YFLogxK
UR - https://www.mendeley.com/catalogue/64117091-4c1b-3689-81f8-e1f6a155b52b/
U2 - 10.1007/978-3-031-16224-4_24
DO - 10.1007/978-3-031-16224-4_24
M3 - Conference contribution
AN - SCOPUS:85140465648
SN - 9783031162237
T3 - Communications in Computer and Information Science
SP - 338
EP - 349
BT - Mathematical Optimization Theory and Operations Research
A2 - Kochetov, Yury
A2 - Eremeev, Anton
A2 - Khamisov, Oleg
A2 - Rettieva, Anna
PB - Springer Nature
T2 - 21st International Conference on Mathematical Optimization Theory and Operations Research , MOTOR 2022
Y2 - 2 July 2022 through 6 July 2022
ER -
ID: 101415278